Skip to content

SETUP.md

Latest commit

 

History

History
55 lines (37 loc) · 2.01 KB

SETUP.md

File metadata and controls

55 lines (37 loc) · 2.01 KB
Raw

One-time setup

Run these from the repository root. They need the GitHub CLI (brew install gh) and one gh auth login.

1. Create the repository and push

The repository lives in the existing Flood-Lab organisation, so there is no organisation to create first.

cd ~/Documents/Python/HydroTuring
gh repo create Flood-Lab/HydroTuring \
    --public \
    --source=. \
    --remote=origin \
    --description "Does your AI hydrologic model conserve mass, energy and momentum?" \
    --homepage "https://flood-lab.github.io/HydroTuring/" \
    --push

2. Turn on Pages

gh api -X POST repos/Flood-Lab/HydroTuring/pages \
    -f 'build_type=workflow'

The site goes live at https://flood-lab.github.io/HydroTuring/ once the pages workflow finishes. Check it with:

gh run watch

3. Topics, so people can find it

gh repo edit Flood-Lab/HydroTuring --add-topic \
    hydrology,benchmark,machine-learning,physics-informed,water-balance,earth-science

4. Labels the contribution flow expects

gh label create accepted        -c 2C7A57 -d "Proposal approved, the probe is yours" -R Flood-Lab/HydroTuring
gh label create probe-proposal  -c 0E6C78 -d "A proposed new probe"                  -R Flood-Lab/HydroTuring
gh label create model-submission -c 3B82F6 -d "A model submitted for evaluation"     -R Flood-Lab/HydroTuring
gh label create good-first-probe -c 7BC4A4 -d "A good place to start"                -R Flood-Lab/HydroTuring
gh label create help-wanted     -c A8501C -d "Unclaimed, we want this"               -R Flood-Lab/HydroTuring

5. Check it worked

gh workflow list -R Flood-Lab/HydroTuring   # probe, model, pages
gh run list      -R Flood-Lab/HydroTuring

The probe workflow runs the acceptance gate. If it is green, the benchmark demonstrably discriminates on GitHub's runners and not only on your laptop.

Afterwards

Ask me to file the fourteen roadmap probes as labelled issues. A markdown list gets read once; issues get claimed.